Crowd Density Research Articles

Multi-scale fusion public gathering recognition based on residual network

With the rapid economic development and high concentration of urban population, people’s income level and quality of life continue to improve, resulting in more and more crowded scenes caused by people going out. Especially in urban commercial centers, transportation hubs, sports venues during important events, tourist attractions, etc., crowd gatherings occur frequently. However, accidents involving crowd gatherings in public places occur frequently, causing heavy casualties and property losses. Therefore, for crowd recognition, this paper proposes a new method to accurately estimate the number of dense crowds. In this method, a density map with accurate pedestrian locations is first generated using the focal inverse distance transform and used as ground truth labels for network training. Then, a multi-scale feature fusion algorithm based on residual network is designed, combining spatial and channel attention mechanisms to improve the accuracy and stability of crowd density estimation. In dense crowds, the phenomenon of overlapping and occlusion of people is very common and serious, making it difficult for existing pedestrian detection methods to distinguish each individual and accurately count the flow of people. To solve this problem, this paper proposes a density map-based method that uses a local maximum detection strategy and a K-nearest neighbor algorithm to convert the density map into the corresponding dense head bounding box. This method can effectively reduce the impact of occlusion and improve the accuracy of people counting. In order to further improve the estimation accuracy, a pattern recognition density peak clustering algorithm is introduced to study the clustered crowds. By treating the head bounding box as an element point, the distance between each element point is calculated, and the density of each point is calculated. Then perform clustering to find the cluster center with the highest density in each class. Finally, by comparing the density of each cluster center with the corresponding density threshold and adopting the corresponding decision-making method, the accuracy of people counting is further improved.

Retail operations strategy for improved customer experience: a better response to crises such as COVID-19 pandemic

Purpose Discount grocery stores (DGSs) are attractive food supply chain (FSC) channels because many cost-conscious Indians use them for monthly needs. Despite capacity, DGSs must address customer concerns about store crowd densities and improve their COVID-19 preparedness. The purpose of this study is to learn how retail operations strategies can improve customer experience and how stores can benefit. Design/methodology/approach The study looked at a case study where retail operations are run more efficiently, and the customer experience is enhanced by standardizing and customizing customer transactions. The potential benefits that customers and retailers might anticipate are then statistically verified. Next, the potential benefits were examined to determine which ones from customers’ and retailers’ views should be prioritized to increase satisfaction. Findings The case situation analysis in the study demonstrates how DGSs can improve their retail operations to reduce customer wait times and provide greater convenience. The study also provides practitioners with potential benefits to pursue from the perspectives of retailers, customers and both retailers and customers. Research limitations/implications This study requires many past transactions and can be considered an extension of the current study, so it does not capture floor space and capacity improvements. Practical implications This research can help FSC retailers compete with upstream supply chain partners and customers in omnichannel retailing. By improving DGS retailer capacity and customer experience, this study can benefit all FSC stakeholders. Originality/value Although there are numerous potential benefits that practitioners can pursue, the current study suggests that practitioners focus on those that can improve retailer and customer satisfaction.

Effect of crowd density, wind direction, and air temperature on the formation of individual human breathing zones in a semi-outdoor environment

This paper presents a comprehensive numerical investigation to predict the human breathing zones (BZs) in crowded semi-outdoor environments. The computational domain consisted of a nine-human block array with integrated nasal cavities subjected to the lower part of the atmospheric boundary layer. Five crowding levels, seven wind directions, and inflow ambient air temperatures (ranging from 10 to 31 °C) were tested to examine the horizontal and vertical formations of the BZs. Validation and verification tests were performed through comparisons with experimental results, a grid independence test, and an evaluation of various randomized distribution scenarios to minimize the uncertainties of the computational fluid dynamics analyses. The horizontal extension of the BZs tripled as the crowding level increased from 0.325 to 4.0 m2/capita. However, the lateral extension was insensitive and remained within 10 cm of the nostrils. Human models can inhale air close to the cheek, neck, and shoulders when an oblique flow is assumed. As the air temperature increased, individuals tended to inhale air from the upper regions, which was influenced by the interrelated thermal properties of the human body. Consequently, under high-temperature conditions, there may be an increased probability of gas-phase contaminant inhalation over greater horizontal distances.

Efficient crowd density estimation with edge intelligence via structural reparameterization and knowledge transfer

Crowd stampedes and related incidents pose significant dangers to public safety and have resulted in numerous fatalities over the past few decades. Estimating crowd density in real-time can help avoid stampedes by providing early warning systems to prevent overcrowding and manage the flow of people. With the increasing prevalence of Internet of Things (IoT), the application of edge computing for field pedestrian density estimation can help to enhance security and efficiency of system. To deal with the issue of crowd scale variation, most previous works rely on heavy backbone networks or complex module, which require high runtime consumption and severely limits the deployment scope of their work. To overcome this issue, we propose multi-branch model Repmobilenet. For multi-scale spatial feature extraction. Repmobilenet equips with lightweight multi-branch depthwise separable convolutional block (DSBlock), which can effectively extract multi-scale feature of dense crowd to cope with scale variation. In the inference phase, the multi-branch structure can be transformed into a single-branch feed-forward structure through structural reparameterization. By this way, Repmobilenet can use multi-branch over-parameterized topology to capture more dense spatial features during training stage and decrease inference latency during inference stage. We also added dilated convolutions in the backend to expand the receptive field of the model. Comparison to state-of-the-art methods, proposed Repmobilenet is able to achieve comparable counting performance while maintaining small model size and low inference latency in ShanghaiTech benchmark. At last, we introduce a multi-layer knowledge distillation method to further increase the model’s fitting capability. By imitating the feature of multiple intermediate layers and final output soft ground truth of the teacher model, the student model can learn compact and efficient knowledge without increasing model size and inference latency. The code can be found in https://github.com/lcxxxiii/Repmobilenet.

Multi-object behavior recognition based on object detection for dense crowds

In recent years, research on multi-object behavior recognition methods based on pose estimation has made progress. However, in dense scenes, multi-object behavior recognition faces problems such as heavy occlusion, many small-scale objects, and an unbalanced distribution of object classes. Meanwhile, multi-object behavior recognition methods based on pose estimation are difficult to adapt to dense scenes. Therefore, we use the idea of object detection to complete the multi-object behavior recognition in dense scenes. However, compared with general object detection, the fine-grained classification and localization constrain each other due to the similarity of object classes, resulting in poor recognition performance. To solve the above problems, we propose a shallow convolutional neural networks (CNNs) module to extract task features separately in localization and classification task parallel branch structure, so that the two vision tasks avoid mutual constraints. Additionally, we propose a feature fusion mechanism to improve the detection performance of small objects by obtaining multi-scale high-level semantic information. To extract balanced and robust features, we exploit the number of objects in an image to generate dynamic class weights and difficulty weights. Experimental results on dense crowds in student classrooms demonstrated that the proposed method achieves superior performance compared to other state-of-the-art methods, with mAP50 of 88.72% and APs of 35.41%.

RGB-D Based Visual SLAM Algorithm for Indoor Crowd Environment

Most current research on dynamic visual Simultaneous Localization and Mapping (SLAM) systems focuses on scenes where static objects occupy most of the environment. However, in densely populated indoor environments, the movement of the crowd can lead to the loss of feature information, thereby diminishing the system’s robustness and accuracy. This paper proposes a visual SLAM algorithm for dense crowd environments based on a combination of the ORB-SLAM2 framework and RGB-D cameras. Firstly, we introduced a dedicated target detection network thread and improved the performance of the target detection network, enhancing its detection coverage in crowded environments, resulting in a 41.5% increase in average accuracy. Additionally, we found that some feature points other than humans in the detection box were mistakenly deleted. Therefore, we proposed an algorithm based on standard deviation fitting to effectively filter out the features. Finally, our system is evaluated on the TUM and Bonn RGB-D dynamic datasets and compared with ORB-SLAM2 and other state-of-the-art visual dynamic SLAM methods. The results indicate that our system’s pose estimation error is reduced by at least 93.60% and 97.11% compared to ORB-SLAM2 in high dynamic environments and the Bonn RGB-D dynamic dataset, respectively. Our method demonstrates comparable performance compared to other recent visual dynamic SLAM methods.

Spectators or Influencers? The Crowd Effect Upon Winning in the NFL: A Natural Experiment

Abstract: Previous research has determined that home field advantage (HFA) is positively related to crowd density. Isolating this effect is a substantial empirical challenge as crowd density is endogenous with home win-likelihood via fan interest. We consider a natural-experimental setting that introduces exogenous crowd-density variation into National Football League (NFL) games. COVID-19 safety protocols allow us to disentangle crowd-presence, crowd-density, and built-environment effects upon HFA. We find strong evidence that crowd presence is a significant, substantial source of HFA, but crowd density is not. No-crowd games in 2020 featured no measurable HFA conditional upon team strengths, whereas partially fan-restricted games featured no significant decline in HFA relative to games without crowd restrictions (2016‒2019 and 2021). Results suggest HFA is fully attributable to crowd presence, with no evidence of stadium familiarity or travel distance effects. Betting markets efficiently predicted the partial-crowd effect from season's outset but adjusted incrementally (behaviorally) to incorporate the true no-crowd effect only by season's end.

Mask focal loss: a unifying framework for dense crowd counting with canonical object detection networks

Mask focal loss: a unifying framework for dense crowd counting with canonical object detection networks

A Wind Protection Design Strategy for Airport Terminal Door Bucket Space Based on Wind Environment Simulations

Airport terminal buildings are large public transportation buildings with complex functions and dense crowds, and the wind environment has significant effects on passenger comfort and the energy consumption of air conditioning systems. In this study, we investigated Xianyang T3 Terminal as an example and used the Phoenics software to conduct comprehensive simulations based on different door bucket forms and parameters, as well as comprehensively considering the impacts of different outdoor wind environments. A terminal door bucket windproof design strategy was proposed based on our results. The results showed that door buckets could effectively reduce the entry of outdoor wind, and built-in door buckets performed the best. The width of the door bucket should be set to 9.3–11.3 m and the depth of the door bucket to 6.7–7.7 m. The height of the door bucket had little impact. A side door should be added to the door bucket, so it can be opened when the outdoor wind speed is high, and a mechanical ventilation system should be introduced to improve the indoor ventilation.

Fourier feature decorrelation based sample attention for dense crowd localization

Crowd localization, which prevails to extract the independent individual features, plays an significant role in critical analysis for crowd scene. Dense trivial features of individual targets are frequently susceptible to interference from complex background features, which makes it difficult to obtain satisfactory predictions for individual targets. Aiming at this issue, a Fourier feature decorrelation based sample attention is proposed for dense crowd localization. The correlation between features are decoupled in the Fourier transform domain, which induces the model to focus more on the true correlation between individual target features and labels. From the perspective of Fourier feature correlation between samples, independence test statistic optimization with cross-covariance operator is developed for feature decorrelation within the sample attention framework. The sample attention with global weight learning is iteratively optimized through matching the prediction loss, which can induce model partial out the spurious correlation between target-irrelevant features and labels. Experimental results show that the method proposed in this paper outperforms the current advanced crowd location methods on public dense crowd datasets.

Pedestrian dynamics on slopes: Empirical analysis of level, uphill, and downhill walking

Pedestrian safety on inclined walking facilities merits greater attention than that on level surfaces. Crowd accidents and domino effects are more likely to occur on slope walkways, especially at high crowd densities, as in the tragic case of the 2022 Itaewon Halloween crowd crush in South Korea. However, experimental investigations of the pedestrian dynamics on slopes with varying inclinations are scarce. In this study, a controlled experiment involving 48 participants was conducted to investigate the differences in pedestrian movement characteristics during level, uphill, and downhill slope walking. A wide range of slope inclinations was considered, namely 0°, 3°, 7°, 12°, 17°, and 22°. The results demonstrated that walking speeds uphill are lower than those on flat terrains, and uphill speeds decrease with increasing inclination. In contrast, increased slope inclinations do not always correspond to reduced downhill speeds. Notably, at inclinations of 3° and 7°, the downhill speed exceeds that of level walking, suggesting that a mild downhill slope can facilitate faster movement. Compared with downhill walking, pedestrians exhibit more lateral motion and significantly lower speeds (p < 0.01) when walking uphill at the same inclination. However, the average flow rate during downhill walking is lower than those observed on flat ground and uphill terrain. This difference may be attributable to the lower pedestrian density during downhill walking to reduce the likelihood of falling. These findings can enhance the understanding of pedestrian dynamics on ramps and have significant implications for the design of safer inclined walking facilities and prevention of potential disasters on such walkways.

Research on emergency evacuation behavior rules of pedestrians under fire and explosion accidents in chemical parks considering social relationships

Chemical parks present a high risk of sudden accidents such as fires and explosions, given the significant number of hazardous sources present. This is why emergency evacuation after accidents is so essential. The behavior of crowds during evacuation is complex, and social relationships among pedestrians can have differentiated effects on the evacuation process. However, most of the current crowd evacuation research has focused on life scenarios and natural disaster contexts. Less attention has been paid to the emergency evacuation of crowds affected by social relations in the context of chemical accident disasters. A comprehensive evacuation model that considers the dynamic evolution of crowd transfer both within and outside the factory area is proposed in this paper. Using the Anylogic software based on the social force model, this study constructs emergency evacuation scenarios for crowd in the context of fire and explosion accidents, and toxic gas leakage incidents. Specifically, this study analyzes the impact of three factors, namely inter-personal distance, group size, and proportion of different group sizes in the total population, on the evacuation efficiency. It has been found that when large groups of pedestrians are split into smaller ones for evacuation, the evacuation efficiency can be effectively improved. The study includes calculations of the lethal range of a toxic gas and simulated the process of relocating large crowds of people to nearby shelters. In addition, a strategy for deploying pedestrian groups based on social relationships was developed for the context of mass evacuation. These measures have the potential to optimize the evacuation efficiency of pedestrian groups in chemical parks, including expanding and modifying shelters, diverting dense crowds from residential plots, using interpersonal spacing of 0.1 m, and splitting large groups of pedestrians into smaller ones. The research results can provide a reference for decision making for large-scale group evacuation in chemical parks, and improve the emergency evacuation capability and local risk response capability in chemical parks.

Construction and application of a simulation optimization model of auxiliary guidance signs in enclosed public places

Many enclosed public spaces with multiple destinations often feature obstacles, such as columns within the building space and dense crowds, which adversely affect the recognition and guidance of pedestrian sign. To mitigate this issue, enhancing traffic efficiency necessitates the implementation of auxiliary guidance signs. However, experience-based schemes for positioning these signs often require iterative adjustments to achieve the desired effect. Thus, this paper endeavors to address this challenge by analyzing the relationship between obstacles and pedestrians’ visual blind areas. It establishes a mathematical model to describe pedestrian visual perception and its correlation with crowd density and individual visual field function. The model employs the effective aggregation area of the population exceeding a certain density threshold as its primary variable and enhances this variable’s function. Subsequently, it formulates an effect function that yields a substantial improvement in various combinations and permutations of auxiliary guidance signs. To exemplify this research, a large subway transfer hall with multiple destinations serves as a simulation model. The study calculates the optimal combination of auxiliary guidance sign placements that align with the enhancement effect function. Based on this combination, an optimized layout plan is derived, resulting in a significant improvement. The results demonstrate that the auxiliary guidance sign layout scheme proposed in this study markedly enhances traffic efficiency in enclosed public spaces. This research serves as a valuable reference for decision-makers seeking to optimize traffic flow in such spaces.

Consistency-Aware Anchor Pyramid Network for Crowd Localization.

Crowd localization aims to predict the positions of humans in images of crowded scenes. While existing methods have made significant progress, two primary challenges remain: (i) a fixed number of evenly distributed anchors can cause excessive or insufficient predictions across regions in an image with varying crowd densities, and (ii) ranking inconsistency of predictions between the testing and training phases leads to the model being sub-optimal in inference. To address these issues, we propose a Consistency-Aware Anchor Pyramid Network (CAAPN) comprising two key components: an Adaptive Anchor Generator (AAG) and a Localizer with Augmented Matching (LAM). The AAG module adaptively generates anchors based on estimated crowd density in local regions to alleviate the anchor deficiency or excess problem. It also considers the spatial distribution prior to heads for better performance. The LAM module is designed to augment the predictions which are used to optimize the neural network during training by introducing an extra set of target candidates and correctly matching them to the ground truth. The proposed method achieves favorable performance against state-of-the-art approaches on five challenging datasets: ShanghaiTech A and B, UCF-QNRF, JHU-CROWD++, and NWPU-Crowd. The source code and trained models will be released at https://github.com/ucasyan/CAAPN.

Automatic Identification of Human Subgroups in Time-Dependent Pedestrian Flow Networks

The study of identifying human subgroups from videos is a significant topic, which has received a lot of attention in multiple disciplines. So far, however, there has been little consideration about combining it with relevant conceptions in network science. Therefore, this paper proposes a novel method for the automatic identification of human subgroups in dynamic pedestrian flows. The spatial proximity and temporal continuity are combined to calculate the interaction intensity between pedestrians, by which a time-dependent pedestrian flow network is constructed. Based on the objective function of weighted partition density, the optimal threshold is used to determine community structures that correspond to human subgroups in frame images. Numerical experiments demonstrate that our method achieves high identification accuracy under various evaluation datasets, and exhibits better performance than existing methods in terms of different crowd densities, various numbers of subgroup members, and certain levels of trajectory noise. Furthermore, this work provides valuable implications for the understanding of subgroup behaviors and the modeling of subgroup movements.

Exploring Ragdoll Physics in Crowd Disaster Simulation

Abstract Objective This study explores the feasibility of applying ragdoll physics, a procedural animation, and related game-engine technologies to create real-time simulations for investigating crowd disasters. Methods We harnessed the power of the Unity Game Engine to develop real-time simulation for the dynamics of a crowd disaster with rendering of scene objects and user interface (UI) elements. Humanoid agents with trigger colliders were linked to color schemes representing impact forces sustained with lethal thresholds ranging from 500 to 550 pounds force for males and 450 to 500 pounds force for females. The simulation enables users to manipulate parameters such as population, crowd density, directional movement and forces applied. Results Our simulation showcases an animated real-time depiction of a crowd disaster and enables visualization of casualties, both injured and dead, based upon a quantitative evaluation of the forces sustained. Conclusions Utilization of the Unity Game Engine for the development of visually-engaging real-time crowd disaster simulation has promising outcomes. This novel approach for modeling and analysis of crowd dynamics may provide valuable insights with applications in mass gatherings, crowd management, safety, and preparedness.

Air Quality in Fitness Centers: Status and Improvement Strategies

As a place where people do physical activity for health, the indoor air quality of fitness center has a significant impact on human health. To systematically understand the current state of indoor air quality in fitness centers, this study reviews recent literature on the indoor environment of fitness centers, analyzes factors that affect air quality, and proposes a series of technical strategies to improve it. The analysis reveals: 1) Significant pollution issues with the indoor air quality in fitness centers; 2) Notable variations in air quality across different functional areas and time periods; 3) Air quality can be improved by adopting appropriate technical measures. This paper recommends that future air quality management strategies should consider factors such as the intensity of physical activity and crowd density to develop ventilation strategies, aiming for targeted air quality management that varies by area and time.

Design of Artificial Intelligence Driven Crowd Density Analysis for Sustainable Smart Cities

Design of Artificial Intelligence Driven Crowd Density Analysis for Sustainable Smart Cities

Personalized virtual reality exposure for panic disorder and agoraphobia: A preliminary neurophysiological study

BackgroundPersonalization is considered an important principle in virtual reality (VR) exposure therapy. We aimed to identify whether personalized VR exposure could provoke increased anxiety in patients with panic disorder and agoraphobia as it is considered the first step in successful treatment for anxiety. MethodsWe performed a double-arm, one-day preliminary study among 28 patients with panic disorder and agoraphobia. Three sessions of VR exposure, including a theater, train, and elevator scenario, were conducted in two groups. In the personalized group (n = 14), the brightness and crowd density were customized based on a pre-assessment. In the control group (n = 14), these conditions were fully randomized. Self-reported anxiety, heart rate, skin conductance, and electroencephalography were measured before, during, and after the VR sessions. ResultsIn the later VR sessions, higher self-reported anxiety levels measured by the Visual Analogue Scale were observed in the personalized exposure group. Increased heart rates during and after the VR sessions were observed in the personalized group. The changes in skin conductance peaks were not significantly different between the groups, but the increase in skin conductance was associated with the participants' perception of presence. The electroencephalogram showed widespread increases in alpha waves in the frontal and temporal areas of the brain in the personalized group than in the control group. ConclusionPersonalized VR exposure elicits stronger anxiogenic effects in patients with panic disorder and agoraphobia as suggested by self-report and neurophysiological data. Personalization of VR exposure has the potential for effective behavioral therapy.

TOWARDS PEDESTRIAN ACCESSIBILITY ANALYSIS IN INDOOR ENVIRONMENTS WITH CROWDS

Abstract. Except in emergency cases, indoor crowds could also decrease the walking speed of pedestrians in daily life. To investigate daily-based accessibility of locations to a pedestrian in the presence of crowds, this paper proposes a method of indoor accessibility analysis. We adopt a hierarchical spatial model to represent indoor environments and evaluate the varied accessibility to a user. The first level refers to a door network representing door connectivity. On the second level, spaces are gridified so that crowd density can be readily measured, and its obstruction to pedestrians can be quantified at different time slots. Instead of generating specific obstacle-avoidance paths in a short period, we depict accessibility variation with two types of paths, i.e., central and periphery paths inside rooms. Based on the hierarchical model, pathways are pre-computed between doors for a user and indexed in the grid model. In a path query, a sequence of doors is derived from the door network, and the pre-computed pathways are readily retrieved to form the two types of paths. Accordingly, walking time can be estimated with crowd density in these grids at each time slot. We conducted an initial test on a building floor to demonstrate the use of the proposed method. We sampled the locations of simulated crowds (50 persons) every 10 seconds, and the time cost between given doors in both paths are compared. A slowdown occurred in the central path, and the periphery path involves the minimum time in the test scenario. In the future, the relationship between crowd density and the possibility of blockage to pedestrians will be further investigated.